Weights & Biases
The near-default ML experiment tracker, now part of CoreWeave Forge
Overview
Weights & Biases' ML/AI developer platform — experiment tracking, model and dataset versioning, and LLM/agent tracing via Weave — became part of CoreWeave Forge on 30 September 2026, alongside OpenPipe and Marimo. Its inference service is now CoreWeave Serverless Inference, an OpenAI-compatible API for open-weight models.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 30 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Experiment tracking
- Artifacts versioning
- Weave basics
- Up to 5 seats
- Up to 10 seats
- 100 GB/mo storage, then $0.03/GB
- 1.5 GB/mo Weave ingest, then $0.10/MB
- Custom pricing
- Extra Weave ingest $0.20/MB
- SSO / governance
What you can produce with Weights & Biases
- Experiment tracking and run comparison via a free, framework-agnostic SDK
- Model and dataset versioning (Artifacts)
- Weave: LLM and agent tracing / observability
- CoreWeave Serverless Inference: OpenAI-compatible API for open-weight models (DeepSeek, Qwen, GLM, Kimi, Nemotron)
- Hyperparameter sweeps
- Shareable reports and dashboards
- Open-source client library (github.com/wandb/wandb)
W&B is close to default infrastructure for ML teams — its experiment-tracking SDK is the sticky part, logging runs, metrics and artifacts with a few lines of code — and as of 30 September 2026 it lives inside CoreWeave Forge, where wandb.ai now redirects. Weave extends it to LLM and agent observability, and W&B Inference has become CoreWeave Serverless Inference, an OpenAI-compatible API for open-weight models on CoreWeave GPUs. CoreWeave's filings put the 2025 acquisition at $1.0 billion, mostly stock. The free tier is genuinely useful with no sales gate; Pro starts at $60 a month for teams under 50, and Enterprise is custom. Watch the usage meters: Weave ingestion is billed at $0.10 per MB on Pro against $0.03 per GB for storage, so heavy tracing can surprise you on the bill. If you train, fine-tune or ship models you'll almost certainly touch it — just expect more CoreWeave branding, and model your ingestion volume first.
What we found
Weights & Biases now sits inside a bigger brand. On 30 September 2026 CoreWeave, which completed its acquisition of W&B in May 2025, folded it into CoreWeave Forge, a platform combining the former Weights & Biases, OpenPipe and Marimo products; wandb.ai now redirects to Forge, and the documentation has moved to CoreWeave's site. The name survives as "Weights & Biases Models", the experiment-tracking product, and the open-source Python client on GitHub remains MIT-licensed and active.
That SDK is still the anchor. A few lines of code log runs, metrics, hyperparameters and artifacts, and the dashboards make comparing training runs and hyperparameter sweeps straightforward; its integrations with common ML frameworks are a recurring reason teams adopt it. Weave, for LLM and agent tracing, now sits alongside a new preview product, Agent Lens, and W&B Inference has become CoreWeave Serverless Inference, an OpenAI-compatible API for open models such as DeepSeek, Qwen, GLM, Kimi and Nemotron.
CoreWeave's annual report puts the purchase consideration at $1.0 billion, mostly in stock — below the roughly $1.7 billion reported when the deal was announced.
Pricing keeps the same shape under the Forge name. The free tier allows up to five seats, 5 GB of storage and 1 GB of Weave ingestion a month; Pro starts at $60 a month for teams under 50 people, with up to ten seats, 100 GB of storage and 1.5 GB of ingestion, then $0.03 per GB of storage and $0.10 per MB of ingestion; Enterprise is custom, with extra ingestion at $0.20 per MB. Academic users get a free Pro licence. Because ingestion is billed per megabyte and storage per gigabyte, heavy tracing can dominate a bill, so model your trace volume before relying on Weave in production.
The free tier remains genuinely useful with no sales gate, which is a large part of why the SDK spread so widely.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Weights & Biases's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Weights & Biases unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Weights & Biases directly →
Spotted something out of date? Suggest an update →
Alternatives to Weights & Biases
More in Code & Dev Tools
